Struct CandleActorCritic 

pub struct CandleActorCritic {
    pub varmap: VarMap,
    /* private fields */
}

Fields

varmap: VarMap

Implementations

impl CandleActorCritic

pub fn new( obs_dim: usize, n_actions: usize, hidden: usize, lr: f64, device: Device, seed: u64, ) -> Result<Self, NNError>

Trait Implementations

impl ActorCritic for CandleActorCritic

fn act(&self, obs: &TensorData) -> Result<ActionOutput, NNError>

Sample actions from the policy (inference, no gradient tracking).

fn value(&self, obs: &TensorData) -> Result<TensorData, NNError>

Compute state values (inference, no gradient tracking).

fn evaluate( &self, obs: &TensorData, actions: &TensorData, ) -> Result<EvalOutput, NNError>

Evaluate the policy on (obs, actions) pairs. Differentiable.

fn ppo_step( &mut self, obs: &TensorData, actions: &TensorData, old_log_probs: &TensorData, advantages: &TensorData, returns: &TensorData, old_values: &TensorData, config: &PPOStepConfig, ) -> Result<TrainMetrics, NNError>

Perform one PPO gradient step. Bundles forward→loss→backward→clip→step.

fn learning_rate(&self) -> f32

fn set_learning_rate(&mut self, lr: f32)

fn save(&self, path: &Path) -> Result<(), NNError>

fn load(&mut self, path: &Path) -> Result<(), NNError>

impl Send for CandleActorCritic

impl Sync for CandleActorCritic